Systematic Mapping of Functional Interactions in Human Brain Dynamics
Poster No:
1756
Submission Type:
Abstract Submission
Authors:
Zhen-Qi Liu1, Andrea Luppi1, Ye Tian2, Ben Fulcher3, Bratislav Misic1
Institutions:
1McGill University, Montreal, Quebec, 2University of Melbourne, Carlton South, Victoria, 3The University of Sydney, Sydney, NSW
First Author:
Co-Author(s):
Introduction:
The human brain consists of anatomically connected regions that exhibit rich functional interactions at the macroscale. Intrinsic functional connectivity (FC), often estimated as zero-lag Pearson correlation between pairs of resting-state fMRI time series, stands as the most popular measure for studying pairwise relationships between brain regions' activities over time. However, this methodological choice is largely arbitrary, and numerous other methods could potentially be used to define FC. Here we profile pairwise functional associations between brain regions using a large library of interaction statistics. We then systematically benchmark multiple features of FC across these measures, including: 1) spatial & geometrical organization; 2) structural & biological associations; 3) fingerprinting and brain–behavior relationships; and 4) information-dynamics flow patterns.
Methods:
Using resting-state functional MRI (rsfMRI) time series of N=326 unrelated subjects from the Human Connectome Project, we adapted the recently developed pyspi toolkit to systematically estimate 239 statistics from 49 pairwise interaction measures across 6 major categories. Group consensus and similarity matrices were calculated. Multimodal neurophysiological networks were used to contextualize the pairwise interactions across 100 cortical regions. Fingerprinting was implemented as the identifiability index, and brain-behavior predictions were carried out using kernel ridge regression with nested cross-validation. Finally, Integrated Information Decomposition (ΦID) was used to understand the specific information flow patterns captured by each pairwise measure.
Results:
The present report broadly surveys pairwise interactions between rsfMRI time series across the neocortex using a library of 239 interaction statistics. We first found that the interaction statistics are highly organized and form meaningful clusters at the group level. Covariance estimators, the most widely used FC measures, are most (anti-)correlated to precision, distance, mutual information, and entropy estimators, confirming the analytical relationship, and most unrelated to spectral estimators. Precision and coherence estimators show the greatest structure–function coupling, while precision and mutual information estimators exhibit higher correlation with multiple inter-regional similarity networks, including gene co-expression and receptor similarity networks. Moving into individual differences, we consider the capacity of pairwise measures to capture individual fingerprints and behavior phenotypes. The precision estimator shows superior performance in uniquely identifying individuals. Covariance, distance, and information-theoretic measures show the greatest predictive accuracy across multiple behavior domains. Finally, we related our findings to finer-scaled information-flow patterns using the novel ΦID framework. We found that commonly used covariance estimators only emphasize the preservation of redundant information ("information storage"); in contrast, some spectral coupling measures also capture unique ("migration") or synergistic ("encryption") information flows.
Conclusions:
In summary, we provide a comprehensive benchmark of pairwise interactions for resting-state haemodynamics in the human brain. We find that many interaction statistics capture distinct aspects of the underlying topological, neurophysiological, and population-level organization. Our analyses reveal some consensus and, at the same time, substantial differences coming from choice of pairwise interaction statistics. Interestingly, precision estimators show greater ability to capture meaningful individual variability, suggesting a relook into classic partial correlation methods. Overall, our project highlights the importance of studying methodological variability and their downstream implications, calling for tailored adoption of analysis methods in both basic and clinical phenotyping studies.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Methods Development
Keywords:
Other - Network Neuroscience;Functional Connectivity;Dynamical System;Information Theory
1|2Indicates the priority used for review
Provide references using author date format
Smith, S. M., Miller, K. L., Salimi-Khorshidi, G., Webster, M., Beckmann, C. F., Nichols, T. E., ... & Woolrich, M. W. (2011). Network modelling methods for FMRI. Neuroimage, 54(2), 875-891.
Pervaiz, U., Vidaurre, D., Woolrich, M. W., & Smith, S. M. (2020). Optimising network modelling methods for fMRI. Neuroimage, 211, 116604.
Suárez, L. E., Markello, R. D., Betzel, R. F., & Misic, B. (2020). Linking structure and function in macroscale brain networks. Trends in cognitive sciences, 24(4), 302-315.
Tian, Y., Margulies, D. S., Breakspear, M., & Zalesky, A. (2020). Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nature neuroscience, 23(11), 1421-1432.
Bazinet, V., Hansen, J. Y., & Misic, B. (2023). Towards a biologically annotated brain connectome. Nature Reviews Neuroscience, 1-14.
Luppi, A. I., Mediano, P. A., Rosas, F. E., Holland, N., Fryer, T. D., O’Brien, J. T., ... & Stamatakis, E. A. (2022). A synergistic core for human brain evolution and cognition. Nature Neuroscience, 25(6), 771-782.